test(kms): keep concurrent rotation regression fail-closed

This commit is contained in:
overtrue
2026-08-02 18:59:11 +08:00
parent 1bafc14a7a
commit 6e26a39893
2 changed files with 8 additions and 53 deletions
+5 -1
View File
@@ -217,11 +217,15 @@ struct Kv2State {
impl Kv2State {
fn new(current_data: serde_json::Value) -> Self {
let mut version_records = BTreeMap::new();
if current_data["baseline_version"].as_u64() == Some(1) {
version_records.insert(1, current_data.clone());
}
Self {
history: BTreeMap::from([(1, current_data.clone())]),
current_data,
current_secret_version: 1,
version_records: BTreeMap::new(),
version_records,
}
}
+3 -52
View File
@@ -696,15 +696,8 @@ impl VaultKmsClient {
async fn ensure_version_history_consistent(&self, key_id: &str, key_data: &VaultKeyData) -> Result<()> {
let recorded = self.recorded_version_numbers(key_id).await?;
// A first-rotation record set (the current v1 record and, at most, the
// next v2 record) can be in flight while another caller is between its
// create-only writes and the baseline CAS. The material check below
// validates the v1 record before adopting it; a later key with a
// missing persisted baseline still indicates a lost baseline and fails
// closed.
if key_data.baseline_version.is_none()
&& let Some(oldest_recorded) = recorded.iter().min().copied()
&& !(key_data.version == 1 && recorded.iter().all(|version| *version <= 2))
{
warn!(
key_id,
@@ -3253,7 +3246,9 @@ mod tests {
use std::sync::Arc;
const ATTEMPTS: usize = 8;
let (vault, client) = scripted_kv2_client(&healthy_key_data()).await;
let mut key_data = healthy_key_data();
key_data.baseline_version = Some(1);
let (vault, client) = scripted_kv2_client(&key_data).await;
let client = Arc::new(client);
let barrier = Arc::new(tokio::sync::Barrier::new(ATTEMPTS));
let tasks: Vec<_> = (0..ATTEMPTS)
@@ -3579,50 +3574,6 @@ mod tests {
);
}
/// A concurrent first rotation may leave both immutable records visible
/// before this caller observes the baseline CAS. That transient state must
/// continue through the create-only/adopt path rather than being reported
/// as a permanently lost baseline.
#[tokio::test]
async fn wired_rotate_recovers_concurrent_first_rotation_without_baseline() {
let key_data = healthy_key_data();
let record_v2 = VaultKeyVersionRecord {
version: 2,
encrypted_key_material: rotated_material(),
created_at: Zoned::now(),
};
let (vault, client) = scripted_client(vec![
ScriptedResponse::ok(kv2_metadata_read_data(1)),
ScriptedResponse::ok(kv2_read_data(&key_data)),
ScriptedResponse::ok(serde_json::json!({ "keys": ["1", "2"] })),
// Another rotation already created the baseline record.
ScriptedResponse::error(400, CAS_CONFLICT_MESSAGE),
ScriptedResponse::ok(kv2_read_version_record_data(&VaultKeyVersionRecord {
version: 1,
encrypted_key_material: key_data.encrypted_key_material.clone(),
created_at: key_data.created_at.clone(),
})),
// Persist the baseline pointer, then adopt the already-created v2.
ScriptedResponse::ok(kv2_write_ack()),
ScriptedResponse::error(400, CAS_CONFLICT_MESSAGE),
ScriptedResponse::ok(kv2_read_version_record_data(&record_v2)),
ScriptedResponse::ok(kv2_write_ack()),
])
.await;
let rotated = client
.rotate_key("wired-key", None)
.await
.expect("an in-flight first rotation must be recoverable");
assert_eq!(rotated.version, 2);
assert_eq!(
parse_write_body(&vault.request_bodies()[8])["data"]["encrypted_key_material"],
serde_json::json!(record_v2.encrypted_key_material),
"the pointer must adopt the immutable v2 material"
);
}
/// The mixed-version corruption path: a node older than versioned rotation
/// performed a lifecycle write, which rewrites the whole key record and
/// silently drops the `baseline_version` it does not know. Version records